The motivation is straightforwardly clinical. Molecular genetic testing remains the gold standard for diagnosing conditions such as sickle-cell disease, beta-thalassemia, BRCA1-linked hereditary cancer syndromes, and cystic fibrosis, but it is expensive, logistically demanding, and unavailable in many low-resource settings. Computational screening tools that flag high-risk patients from routinely collected measurements—complete blood count indices, inflammatory markers, targeted gene-expression assays—could triage patients toward confirmatory testing earlier and more cheaply. Deep neural networks have been proposed widely for this purpose because they can learn complex, non-linear relationships among biomarkers without hand-engineered diagnostic rules. Yet most such tools produce a single softmax point estimate that is often overconfident, particularly for atypical inputs or cases lying in regions of genuine class overlap. In a diagnostic setting, an overconfident wrong answer is arguably more dangerous than a low-confidence one, because it removes the clinician’s incentive to order the confirmatory test that would catch the error.

The team, led by V. S. Keerthana and Tarak Hussain, trained a feed-forward multilayer perceptron with three hidden layers of 128, 64, and 32 units, each stabilized by batch normalization and regularized by dropout with a rate of 0.3. The network classified 1,000 patient records into five genetic-disease-associated groups using twelve clinical and molecular biomarkers, including hemoglobin concentration, fetal hemoglobin fraction, serum ferritin, sweat chloride, the proportion of sickled red blood cells, BRCA1 expression, and TP53 mutation status. Class distribution was nearly balanced, with 196 to 205 samples per category, avoiding the confound of extreme class imbalance. Performance was assessed with stratified five-fold cross-validation against logistic regression, random forest, support vector machine, and gradient boosting classifiers, all running on the same standardized features.

The accuracy results were humbling for the deep learning camp. The neural network achieved 94.30 percent cross-validated accuracy, plus or minus 1.36 percent—numerically lower than all four classical baselines, which ranged from 94.90 to 95.50 percent. An exploratory McNemar’s test on paired predictions suggested the gaps against random forest (95.50 percent, p = 0.0018) and support vector machine (95.30 percent, p = 0.0063) crossed the threshold of statistical significance, while logistic regression and gradient boosting were statistically indistinguishable from the network. The authors are careful to note that their pooled-fold significance test violates the independence assumption McNemar’s test requires, since cross-validation folds share overlapping training sets, so these p-values should be read as suggestive rather than confirmatory. Still, the asymmetry in paired disagreement counts was striking: across all five folds, the network corrected only one test sample that random forest got wrong, while random forest corrected thirteen samples the network misclassified. The finding aligns with a growing body of evidence that tree-based and kernel methods remain highly competitive with deep learning on structured tabular data of this size.

Where the deep learning approach pulled ahead was in uncertainty. The technique at the center of this advantage, Monte Carlo dropout, exploits the fact that dropout—normally a training-time regularizer that randomly silences a fraction of neurons—can be kept active at inference time. Instead of one deterministic forward pass, the network runs up to 200 stochastic passes per input, each with a different random pattern of dropped units, producing a collection of softmax vectors. The mean of these vectors serves as the final predicted probability distribution, while the Shannon entropy of that mean quantifies total predictive uncertainty. Crucially, this total decomposes into two components with different clinical meanings: aleatoric uncertainty, the average of the per-pass entropies, reflects ambiguity inherent to the data itself, such as borderline biomarker presentations; epistemic uncertainty, computed as the mutual information between the mean prediction and the per-pass entropies, reflects model doubt that could in principle be reduced with more or more representative training data. In this dataset, aleatoric uncertainty dominated, accounting for 69.9 percent of total predictive entropy—suggesting most low-confidence predictions reflect genuine biological overlap between disease classes rather than a fixable data deficiency.

The uncertainty estimates proved convergent and usable. Accuracy and mean predictive entropy stabilized by 50 to 100 stochastic passes, with the bootstrap standard deviation of the entropy estimate falling nearly 5.5-fold from 0.0123 nats at 10 passes to 0.0022 nats at 200. The estimates also tracked prediction quality: when held-out test predictions were sorted into low, medium, and high uncertainty terciles by predictive entropy, the error rate roughly doubled from 6.0 percent in the lowest tercile to 12.0 percent in the highest, and 60 percent of all misclassifications landed in the high-uncertainty group despite that group containing only a third of samples. The conditional probability of high uncertainty given an incorrect prediction, 0.600, was nearly twice the 33 percent base rate expected if uncertainty carried no information about correctness. Incorrect predictions, as a population, were shifted toward substantially higher entropy than correct ones.

The most immediately practical result came from selective prediction, sometimes called abstention analysis. By ranking held-out predictions in increasing order of predictive entropy and discarding the most doubtful ones, the effective accuracy of retained predictions rose sharply: removing the most uncertain 20 to 30 percent of predictions pushed retained-sample accuracy above 95 percent, up from the full-coverage figure of 93.33 percent. The area under the risk-coverage curve, 0.0662, summarizes this gain in a single statistic. The authors frame this as a candidate decision-support mechanism—automatically accepting confident predictions while referring uncertain ones for physician review or confirmatory genetic testing—rather than an autonomous diagnostic tool. The computational cost of the approach is negligible in absolute terms: the network has 12,165 trainable parameters, trains in about half a second on a standard CPU, and Monte Carlo inference with 100 passes on the entire 150-sample test set took under 50 milliseconds, roughly 121 times the deterministic cost but still trivial for a screening application that is not latency-critical.

Calibration told a more nuanced story. On the held-out test partition, the Monte Carlo dropout model achieved 93.33 percent accuracy with macro F1 of 0.933, macro ROC-AUC of 0.971, and Matthews correlation coefficient of 0.917. But its expected calibration error of 0.076 was numerically the worst among the six model and inference configurations evaluated, exceeding all four classical baselines, which ranged from 0.024 for the support vector machine to 0.066 for gradient boosting. Yet Monte Carlo averaging substantially reduced the negative log-likelihood relative to deterministic inference, 0.370 versus 0.445—a metric that penalizes overconfident wrong predictions far more heavily than bin-wise calibration measures do. The authors interpret this as evidence that no single calibration metric tells the whole story: the benefit of Monte Carlo dropout is concentrated in limiting the damage of confident mistakes rather than in aligning confidence with accuracy across probability bins. An ablation study across seven dropout configurations revealed a clear monotonic trade-off, with lower dropout rates yielding better-calibrated but less richly quantified uncertainty, and higher rates producing poorer calibration but larger, more clearly separated uncertainty estimates.

Robustness testing added further confidence, with one instructive surprise. Accuracy was consistent across five random initialization seeds at 93.07 percent plus or minus 0.33 percent, and held up even when training data was cut to a quarter of available samples. When zero-mean Gaussian noise was injected into test features at inference time, accuracy degraded sharply beyond a noise standard deviation of 0.5 standardized units, falling from 93.33 percent to 69.33 percent at 1.0 and 59.33 percent at 2.0—but mean predictive entropy rose monotonically and substantially across the same range, from 0.522 to 0.840 nats. This co-movement of increasing uncertainty with degrading accuracy is exactly the property a trustworthy uncertainty estimate should have under distribution shift: the model becomes appropriately less confident on corrupted inputs rather than remaining artificially assured. The surprise came from the training-data experiment, where mean predictive entropy was lowest, not highest, when the model trained on the smallest subset—contrary to the naive expectation that less data should produce more epistemic uncertainty. The authors present this pattern in full and flag it as an open question for multi-seed replication rather than burying it.

Every result in the study rests on internal computational validation of a single publicly available dataset of 1,000 records whose real-world clinical versus synthetic provenance could not be independently confirmed, and no external or prospective clinical testing was performed. The authors also emphasize that the composite five-class formulation—spanning conditions that would never share a differential-diagnosis pathway at the bedside—is a deliberately heterogeneous methodological benchmark for stress-testing the uncertainty pipeline, not a proposed simultaneous diagnostic instrument; a real deployment would use separate, condition-specific models triggered by the marker subset clinically indicated for a given presentation. Permutation importance analysis showed the classification signal was governed mainly by direct molecular and hematological markers, with sickled red blood cell percentage and BRCA1 expression jointly carrying most of the discriminative ability, while age, gender, and family history contributed essentially nothing once those direct markers were available. The broader lesson for the field is twofold: do not default to deep learning for structured genetic disease data without empirical validation, since classical algorithms matched or beat it here under standard, non-tuned hyperparameters; but recognize that the deep learning approach’s real advantage was not accuracy at all, rather the ability to know what it did not know—a capacity the authors argue is worth more than a single percentage point of accuracy in any screening workflow where a confident error can delay the right test.

Subject of Research: Machine learning

Article Title: Deep learning-based genetic disease prediction with Monte Carlo (MC) dropout uncertainty estimation: a comparative study with classical machine learning baselines

Article References: Keerthana, V. S., & Hussain, T. (2026). Deep learning-based genetic disease prediction with Monte Carlo (MC) dropout uncertainty estimation: a comparative study with classical machine learning baselines. Frontiers in Genetics, 17, Article 1969295. https://doi.org/10.3389/fgene.2026.1969295

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1969295

Keywords: deep learning, Monte Carlo dropout, uncertainty quantification, genetic disease prediction, machine learning benchmarking, model calibration, clinical decision support, precision medicine, Deep, learning-based, genetic, disease

The research, led by Galang Adira Prayoga of Institut Teknologi Bandung together with Emir Husni, Reza Darmakusuma and Herto Dwi Ariesyady, introduces a framework called Real-Time Sensing and Visual Information-Guided Physics-Informed Neural Networks, or RTS-VG PINNs. The core insight is that turbidity and the degree of flocculation, the two indicators most commonly used to judge how well coagulation is working, are inherently lagging signals. They measure the macroscopic outcome of particle aggregation and sedimentation, processes that take time to propagate from the bottom of a thickener up to the clear supernatant layer where sensors operate. By the time a change is visible in turbidity, the underlying process has already moved on, forcing conventional control strategies into reactive adjustments based on stale information.

To get ahead of that lag, the team combined two sources of information that are usually kept separate: continuous water quality sensing and live video of the flocculation process. On the vision side, the framework uses two complementary convolutional neural networks. A ResNet-50 based two-stage detector provides high-resolution feature extraction, learning hierarchical representations that range from low-level textures to high-level semantic descriptions of floc structure, with residual connections helping the deep network stay sensitive to subtle changes in floc morphology and density. In parallel, a YOLOv8 one-stage detector performs feature extraction and object detection in a single pass, dramatically reducing computational latency so that floc evolution can be identified and tracked in real time.

The camera setup itself was carefully engineered. A Canon 250D recorded video at 60 frames per second at 1920 by 1080 pixel resolution, positioned in front of a jar test reactor under controlled lighting to minimize reflections. Each experimental run followed a standardized protocol of two minutes of rapid mixing at 200 rpm, ten minutes of slow mixing at 30 rpm, and five minutes of settling. Regions of interest were defined within the observation window to focus the networks on representative floc behavior while suppressing background noise, and the resulting image sequences were synchronized with calibrated turbidity, pH and electrical conductivity measurements through a central computing system.

What makes the framework genuinely distinctive is how the visual information is used. Rather than treating the camera output as a standalone classifier, the researchers map the extracted floc features onto a physically meaningful parameter: the time-varying effective sedimentation velocity, v0(t). This velocity dynamically governs the Kynch sedimentation partial differential equation, a classical description of how particle concentrations evolve in space and time as suspensions settle. A small subnetwork called ParamNet learns the nonlinear relationship between image-derived features and this velocity, so the physics model continuously adapts to the evolving coagulation state instead of assuming fixed settling behavior.

The neural network at the heart of the system takes time and depth coordinates as inputs and outputs a predicted floc concentration field, effectively reconstructing the internal density distribution of the suspension, something no external sensor can measure directly. Training is governed by a composite loss function that balances four competing demands: a physics loss that penalizes violations of the Kynch equation, a data loss that anchors predictions to experimental observations, an initial condition loss, and a boundary condition loss enforcing a no-flux constraint at the top surface. The result is a model that is not merely fitting curves but adhering to conservation of mass and sedimentation dynamics, with physics-based residuals converging to the order of ten to the minus three.

The inverse problem is where the framework earns its practical value. A second network takes initial operating conditions, including pH, temperature, velocity gradient, initial zeta potential and initial turbidity, and predicts how zeta potential, turbidity, floc size and settling velocity evolve as functions of coagulant dose. Crucially, these predictions are constrained by embedded physical laws: a Langmuir-type adsorption relation captures the nonlinear surface charge response, while a Smoluchowski-style aggregation and breakup balance governs turbidity evolution. The model can then infer the dose that minimizes final turbidity, benchmarked against an independent experimental ground truth defined as the dose achieving the lowest measured supernatant turbidity in standardized jar tests.

Experiments spanned natural river water from a PDAM utility and artificial kaolin suspensions prepared from tap water and distilled water, with initial turbidities around 50 to 60 NTU, pH from 6.2 to 7.8, and zeta potentials from minus 3.1 to 0.3 mV. Polyaluminum chloride doses ranged from 12 to 120 mg/L, with the most favorable flocculation consistently observed between 24 and 84 mg/L. Against conventional benchmarks including Random Forest, Gradient Boosting, Support Vector Machine and Multilayer Perceptron models, RTS-VG PINNs achieved the highest coefficient of determination, approximately 0.498 compared with 0.484 for Random Forest, along with the lowest RMSE and MAE. The authors are candid that this improvement is modest and, given only ten independent experimental runs, should be read as evidence of potential benefit from combining visual information and physical constraints rather than statistically significant superiority.

The framework also incorporates Monte Carlo Dropout as an approximate Bayesian method for uncertainty quantification. By keeping dropout layers active during inference and running multiple stochastic forward passes, the system produces a distribution of possible dose estimates rather than a single number. The predictive mean serves as the recommended dose, while the predictive variance flags conditions where the model is uncertain, such as unfamiliar water chemistry or ambiguous visual flocculation states. High-variance predictions could be routed to operators for verification before any automated dosing action, a safety layer the authors emphasize is a decision-support indicator for future online implementation rather than a validated closed-loop control criterion.

The researchers are equally transparent about the limits of the current work. The one-dimensional Kynch formulation cannot capture the three-dimensional hydrodynamics of real treatment basins, the validation relied on only two independent laboratory-scale validation runs, and no end-to-end closed-loop dosing experiment was performed, leaving inference latency and actuator response unquantified. Future directions include coupling the vision-informed framework with computational fluid dynamics, incorporating population balance models to resolve microscale aggregation, and extending validation to pilot and full-scale plants under seasonal variation. Even so, the study demonstrates something conceptually important: that a camera watching particles clump together, fused with classical sedimentation physics and live sensor data, can anticipate what a turbidity meter will only report minutes later. If that promise survives scale-up, the humble jar test may finally gain a predictive successor.

Subject of Research: Physics-informed neural networks with computer vision for real-time coagulant dose optimization in water treatment

Article Title: Visual information guided physics informed neural networks framework for predictive coagulant dose optimization

Article References: Prayoga, G. A., Husni, E., Darmakusuma, R., & Ariesyady, H. D. (2026). Visual information guided physics informed neural networks framework for predictive coagulant dose optimization. Discover Artificial Intelligence, 6(1), Article 1380. https://doi.org/10.1007/s44163-026-02372-z

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02372-z

Keywords: physics-informed neural networks, coagulation, water treatment, computer vision, flocculation, Kynch sedimentation, YOLOv8, ResNet-50, coagulant dose optimization, turbidity sensing, Monte Carlo dropout, machine learning

The core idea behind the framework is a shift in what the neural network is trained to output. Standard deterministic classifiers produce a single probability vector, typically normalized through a softmax function, which describes the likelihood of each class but says nothing about the reliability of that estimate. Evidential deep learning takes a different route: instead of predicting class probabilities directly, the network predicts the parameters of a higher-order distribution over those probabilities. In this work, the authors place a Dirichlet distribution over the class probabilities of a binary stroke prediction task. The Dirichlet distribution acts as a probability distribution over probability distributions, and its parameters can be interpreted as evidence collected by the network in favor of each class. From this single higher-order output, the model can compute both a point estimate of the prediction and a full characterization of how uncertain that prediction is, all in one forward pass through the network.

What makes this approach especially valuable for clinical decision support is the way it decomposes uncertainty into two fundamentally different types. Epistemic uncertainty, sometimes called model uncertainty, reflects gaps in the model’s knowledge: cases that look unfamiliar, ambiguous or unlike anything the network saw during training. Aleatoric uncertainty, by contrast, captures noise inherent in the data itself, such as overlapping symptom profiles or measurement variability that no amount of additional training data could resolve. The evidential formulation allows both components to be estimated explicitly from the Dirichlet parameters. This distinction matters enormously in practice. High epistemic uncertainty flags cases where the model is essentially guessing and where a clinician should look more carefully, while high aleatoric uncertainty signals genuinely difficult cases where even expert disagreement would be expected. The study’s uncertainty analysis confirmed that ambiguous and misclassified cases indeed exhibited higher epistemic uncertainty, meaning the model conveniently identified its own weak spots as low-confidence predictions.

The performance numbers reported in the paper are striking. Across distinct experimental repeats, the model achieved an area under the receiver operating characteristic curve approaching 0.99, together with high average precision, indicating near-perfect separation between stroke and non-stroke cases. At the operating point selected using Youden’s J statistic, a standard method for balancing sensitivity and specificity, the model reached a sensitivity of 0.968, a specificity of 0.949 and a negative predictive value of 0.980. In clinical terms, this means the system correctly identified nearly 97 percent of true stroke cases while correctly clearing almost 95 percent of non-stroke cases, and when it predicted that a patient had not suffered a stroke, that assessment was correct 98 percent of the time. For a condition in which every minute of delayed diagnosis destroys neurons at an staggering rate, a reliable screening tool with this profile could meaningfully accelerate triage.

Just as important as raw discrimination is calibration, the alignment between predicted probabilities and observed outcomes. A model may rank cases correctly yet be systematically overconfident, reporting 99 percent certainty on cases where it is wrong one time in five. The calibration analysis in the study demonstrates that the predicted probabilities produced by the evidential framework lie closer to the observed outcomes than those of conventional approaches, meaning that when the model says it is 90 percent confident, it is right roughly 90 percent of the time. Well-calibrated confidence estimates are precisely what safety-critical clinical applications require, because they allow downstream users, whether physicians or automated triage protocols, to set thresholds and escalation rules that behave predictably in the real world.

To establish that the full evidential formulation was genuinely responsible for these gains, the authors conducted comprehensive ablation studies comparing their approach against several established baselines. These included standard convolutional networks trained with conventional losses, networks trained with focal loss, a technique designed to down-weight easy examples and focus learning on hard ones, and Monte Carlo dropout, the most widely used uncertainty estimation method in deep learning. The proposed full evidential formulation consistently outperformed all of these alternatives, both in discriminative performance and in the quality of its uncertainty estimates. This systematic comparison strengthens the case that the Dirichlet-based approach is not merely a cosmetic addition but a substantive improvement in how the network reasons about its own predictions.

One of the most compelling practical findings concerns computational cost. Monte Carlo dropout estimates uncertainty by running the network through many stochastic forward passes, each with different units randomly deactivated, and aggregating the results. In this study, MC Dropout required 10 stochastic forward passes per prediction to produce its uncertainty estimates. The evidential framework, by contrast, required only a single forward pass, because the Dirichlet parameters encode the uncertainty information directly. That represents an order-of-magnitude reduction in inference cost for uncertainty-aware prediction, a difference that matters when models must run at scale in hospital information systems, on resource-constrained hardware, or in time-critical emergency settings where stroke triage decisions cannot wait for repeated computation.

The clinical context of the work adds weight to its technical contributions. The research team includes neurologists from the Division of Neurology at Chulalongkorn University’s Faculty of Medicine and the Chulalongkorn Stroke Center at King Chulalongkorn Memorial Hospital, ensuring that the framework was developed with direct awareness of the realities of stroke care. Stroke remains one of the leading causes of death and long-term disability worldwide, and the therapeutic window for interventions such as thrombolysis and thrombectomy is measured in hours. Decision support systems that can rapidly and reliably flag likely stroke cases, while transparently signaling when a case falls outside their competence, fit naturally into triage workflows where the cost of a missed case is catastrophic and the cost of unnecessary escalation is delay for other patients.

The broader significance of the study extends beyond stroke. The pattern it illustrates, in which high discriminative performance is paired with explicit, decomposed, well-calibrated uncertainty quantification at low computational cost, is a template for deploying deep learning in any safety-critical domain. The authors note that high discriminative performance alone does not ensure reliable clinical decision support, and their results back that claim empirically: the evidential model not only matched or exceeded baseline accuracy but also produced uncertainty signals that correctly tracked ambiguity and error. As regulatory frameworks for medical artificial intelligence increasingly demand transparency about model limitations, methods that quantify what a model does not know are likely to move from research curiosity to deployment requirement.

Funded by the Thailand Science Research and Innovation Fund, the Ratchadapisek Somphot Fund, Chulalongkorn University’s Second Century Fund and the National Science, Research and Innovation Fund through the Program Management Unit for Human Resources and Institutional Development, the work reflects a growing international effort to make machine learning trustworthy enough for the bedside. The study, published open access on 9 October 2026 with a permanent DOI, demonstrates that a single neural network can, in one pass, deliver near-perfect stroke discrimination, honest confidence estimates, a clear separation of what it knows from what it merely guesses, and the computational efficiency to do it all in real time. For a field where the difference between a confident error and a flagged uncertainty can shape a patient’s outcome, that combination may prove to be the most important diagnostic result of all.

Subject of Research: Uncertainty-aware evidential deep learning for reliable stroke detection and clinical decision support

Article Title: Uncertainty-aware evidential deep learning for reliable stroke detection and clinical decision support

Article References: Saleem, M. A., Javeed, A., Akarathanawat, W., Chutinet, A., Suwanwela, N. C., Chaitusaney, S., Benjapolakul, W., & Kaewplung, P. (2026). Uncertainty-aware evidential deep learning for reliable stroke detection and clinical decision support. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02540-9

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02540-9

Keywords: stroke detection, evidential deep learning, uncertainty quantification, Dirichlet distribution, epistemic uncertainty, aleatoric uncertainty, calibration, clinical decision support, machine learning, neural networks, Monte Carlo dropout, computational efficiency

The core challenge the researchers set out to solve is one that plagues nearly every medical prediction task. In real-world health datasets, patients with hypertension are vastly outnumbered by those without it, which biases standard classifiers toward the majority class and hides the very cases doctors most need to find. Rather than simply duplicating minority samples, the team employed an advanced synthetic sampling strategy built on distribution-aware weighting mechanisms. This approach generates representative minority-class samples that respect the underlying structure of the feature space, reducing majority-class bias while improving how minority instances are represented. The result is a balanced training foundation on which the three proposed models could be built and fairly compared.

The first architecture, dubbed HighCeptNet, fuses two well-established deep learning ideas into a single pipeline. Highway networks use gated skip connections that allow information to flow unimpeded through very deep stacks of layers, mitigating the vanishing gradient problems that once limited network depth. Inception networks, by contrast, process the same input through parallel filters of different sizes, capturing patterns at multiple scales simultaneously. By combining deep residual learning with multi-scale feature extraction, HighCeptNet gains a sharper discriminative capacity than either component alone, allowing it to separate subtle pre-hypertensive signals from background noise in patient data.

The second model, HighCeptALNet, tackles a different bottleneck: the cost of expert annotation. Labeling medical records requires clinician time, which is scarce and expensive. The researchers incorporated uncertainty-based active learning powered by Monte Carlo dropout, a technique in which dropout layers remain active at inference time so the network produces a distribution of predictions for each input rather than a single fixed output. The variance across those repeated stochastic passes serves as an estimate of the model’s uncertainty. Instances with the highest uncertainty are flagged as the most informative and prioritized for labeling, while confidently classified samples are left alone. This selective sampling substantially reduces labeling costs while preserving high predictive performance, a trade-off that could prove decisive for hospitals hoping to deploy such systems at scale.

The third and most successful architecture, FireCeptNet, adds a layer of nature-inspired intelligence to the training process itself. Instead of relying on gradient-based tuning or manual hyperparameter search, which the authors note often traps models in local optima, slows convergence, and inflates computational cost, FireCeptNet employs the firefly algorithm. This metaheuristic mimics the way fireflies use light flashes to attract one another, with candidate solutions moving toward brighter, better-performing neighbors across the search space. By adaptively tuning hyperparameters this way, the model explores the optimization landscape more thoroughly, improving training stability, convergence behavior, and robustness across diverse clinical settings.

The performance gains reported are striking. Compared against existing and baseline methods, HighCeptNet achieved relative improvements of 3.30 percent in accuracy, 3.33 percent in F1-score, 3.30 percent in recall, 3.26 percent in precision, 10.13 percent in the Matthews Correlation Coefficient, 11.39 percent in Cohen’s kappa, and 28.95 percent in ROC-AUC. HighCeptALNet pushed further, with gains of 6.59 percent in accuracy, 7.78 percent in F1-score, 7.69 percent in recall, 3.26 percent in precision, 16.46 percent in MCC, 17.72 percent in Cohen’s kappa, and the same 28.95 percent jump in ROC-AUC. FireCeptNet delivered the strongest results of all: 7.69 percent in accuracy, 7.78 percent in F1-score, 7.69 percent in recall, 5.43 percent in precision, 21.52 percent in MCC, 20.25 percent in Cohen’s kappa, and 28.95 percent in ROC-AUC, consistently outperforming every baseline and the other proposed architectures under diverse optimization settings.

What distinguishes this work from many high-performing machine learning papers is the rigor of its evaluation. Rather than relying on a single train-test split, the team implemented 10-fold cross validation, rotating the data through ten partitions to ensure the results were not an artifact of a lucky split. They complemented this with confidence interval estimation for every reported metric, giving readers a sense of the plausible range of each score, and validated statistical significance using the Mann–Whitney U test, a non-parametric method suited to comparing model performance distributions. Together, these measures suggest the improvements are genuine rather than statistical noise, a standard of evidence that clinical deployment ultimately demands.

Perhaps most importantly for real-world medicine, the framework does not hide its reasoning. The researchers applied SHapley Additive exPlanations, or SHAP, at two levels. Globally, Shapley additive attribution reveals which features drive the model’s predictions across the entire dataset, offering population-level insight into the risk factors the algorithm considers most decisive. Locally, instance-level interpretable attribution explains why a specific patient received a specific prediction, breaking the output into contributions from individual inputs. This dual transparency is designed to build clinical-level trust, allowing physicians to audit the model’s logic rather than accept or reject it blindly.

The authors argue that their framework fills a persistent gap between predictive accuracy, annotation efficiency, and clinical interpretability, three goals that are usually pursued separately. By addressing class imbalance, cutting labeling costs through uncertainty-guided active learning, stabilizing training with firefly optimization, and opening the black box with SHAP-based explanations, the study presents a statistically stable and interpretable solution for early hypertension prediction. Funded by Prince Sattam bin Abdulaziz University and published open access, the research offers a template that could extend well beyond hypertension, pointing toward a future in which medical AI systems are not only accurate but also honest about what they know, what they do not, and why they decided.

Subject of Research: Explainable deep learning frameworks for early hypertension prediction

Article Title: An explainable framework for hypertension prediction using uncertainty-based Monte Carlo dropout and firefly optimization

Article References: Chihaoui, N., Javaid, N., Pamucar, D., Alotaibi, S., Aldosari, H., & BinJwair, A. (2026). An explainable framework for hypertension prediction using uncertainty-based Monte Carlo dropout and firefly optimization. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02357-6

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02357-6

Keywords: hypertension prediction, deep learning, Monte Carlo dropout, active learning, firefly optimization, class imbalance, SHAP, explainable AI, 10-fold cross validation, Mann-Whitney U test, Highway network, Inception network

The research, led by Mohammad Rasoul Sedaghat Rad and Atabak Feizi of the University of Mohaghegh Ardabili together with Hosein Alizadeh of Iran University of Science and Technology, focuses on the Balikhlouchai River watershed in Ardabil Province, northwestern Iran. The team assembled a quarter-century of daily observations spanning 1999 to 2024, combining streamflow records from the Ardabil Regional Water Company with rainfall and mean temperature data from the Iran Meteorological Organization. That 26-year record gave the models enough examples of seasonal cycles, droughts, and flood episodes to learn the watershed’s behavior — but the investigators deliberately refused to let the flood peaks drown in the statistical noise of the thousands of routine low-flow days that dominate any daily hydrological series.

To characterize what counts as a peak, the researchers turned to the statistics of extremes. Flow conditions above the 95th percentile of the record were defined as upper-tail peak events, a threshold-based definition that isolates the rarest and most dangerous 5 percent of daily flows. This choice matters because standard training procedures implicitly optimize for the bulk of the data; a model can achieve impressive overall accuracy while systematically missing every major flood. The team therefore designed a two-stage training strategy. In the first stage, models were pre-trained with the Huber loss, a robust objective that behaves like squared error for small mistakes and like absolute error for large ones, damping the influence of outliers while the networks learn general flow dynamics. In the second stage, the models were fine-tuned with a peak-weighted mean squared error that explicitly amplifies the penalty for errors during high-flow conditions, forcing the networks to allocate their capacity to the events where a wrong answer carries real consequences.

The architectural search was unusually broad. The team benchmarked baseline models against a family of hybrid deep learning designs, including combinations of convolutional neural networks with recurrent layers, U-shaped encoder–decoder architectures borrowed from image segmentation, attention-based models, and Transformer-based designs of the kind that have revolutionized language processing. Convolutional layers act as feature extractors, scanning windows of past rainfall, temperature, and flow to detect local patterns such as storm signatures; recurrent layers such as long short-term memory units and gated recurrent units carry information across time steps, retaining the memory of antecedent conditions that govern how a watershed responds to new rain. Bidirectional variants read the input sequence in both directions during training, while the U-shaped encoder–decoder structure, adapted from the U-Net architecture originally developed for biomedical imaging, compresses the input into a compact representation and then reconstructs it, allowing the network to blend fine-scale temporal detail with broader context.

When the dust settled on the deterministic benchmarks, one family stood out. The U-shaped CNN–RNN hybrids delivered the most balanced performance for peak flow forecasting, particularly on the operational diagnostics that measure whether a model actually detects flood events rather than merely tracking average flow. The standout was U-CNN-BiGRU, a U-shaped hybrid pairing convolutional feature extraction with a bidirectional gated recurrent unit, which achieved the best integrated rank with a multi-criteria decision analysis score of 0.69. That score reflected a rare combination of qualities: strong overall accuracy, a low tendency to raise false alarms, and favorable behavior in the trade-off between how confident the model’s uncertainty estimates were and how reliable those estimates proved in practice. Its sibling, U-CNN-BiLSTM, earned a distinction of its own, showing the highest robustness under noisy meteorological inputs among all the models that consumed weather data — a property that matters enormously in operational settings, where forecasts are fed imperfect real-world observations rather than pristine historical records.

What elevates the study beyond a model horse race is its treatment of uncertainty. The authors separated two fundamentally different kinds of doubt. Epistemic uncertainty — the model’s own lack of knowledge — was quantified using Monte Carlo Dropout, a technique in which dropout layers are kept active at prediction time so that the network is effectively sampled many times, producing a distribution of forecasts whose spread reveals where the model is unsure. Aleatory robustness, by contrast, concerns the irreducible noise in the inputs themselves, and was probed by perturbing the meteorological inputs and observing how violently the predictions responded. A forecast that looks precise but swings wildly when rainfall is jittered by a small amount is a dangerous forecast; the robustness analysis exposed exactly which architectures could be trusted when the weather data wobbled.

Interpretability received equally rigorous treatment. Using SHAP — Shapley Additive Explanations, a game-theoretic method that attributes each prediction to its contributing inputs — the team performed peak-conditioned analyses to discover what the winning models were actually looking at when floods loomed. The answer was striking: short-lag antecedent streamflow, above all the flow recorded just one day earlier, dominated the one-day-ahead predictions. Rainfall and temperature played a secondary, event-dependent role, refining the forecast rather than controlling it. In other words, the rivers of this watershed carry so much memory in their channels and soils that yesterday’s flow is the single best clue to tomorrow’s, with the weather sharpening the picture when a storm is actually in play. For hydrologists, this is a reassuring sign that the networks learned physically sensible behavior rather than exploiting spurious correlations — and for forecasters, it identifies precisely which measurements must never fail in an operational system.

The final layer of the framework is perhaps its most distinctive contribution: model selection by multi-criteria decision analysis. Rather than crowning a winner on a single accuracy metric, the researchers integrated deterministic performance, peak-detection reliability, uncertainty behavior, and interpretability into a structured MCDA ranking. This approach acknowledges a truth that operational agencies know well — a model that is marginally less accurate but far more honest about its uncertainty and far less prone to false alarms may be worth far more in a flood early-warning context than a leaderboard champion. The framework as a whole, the authors argue, can serve as a predictive component within data-driven early-warning systems, giving decision-makers not just a number but a calibrated, explained, and stress-tested one.

The implications reach well beyond one Iranian watershed. As climate change intensifies the hydrological cycle, extreme flows are becoming less predictable from historical patterns alone, and the gap between average skill and peak skill is widening into a genuine safety hazard. Studies across basins from the Amazon to the Qinghai-Tibet Plateau have documented how deep learning models transform streamflow prediction, yet the field has lacked a disciplined recipe for making those models peak-aware, uncertainty-aware, and interpretable at once. By combining a two-stage loss strategy, tail-focused diagnostics, Monte Carlo uncertainty separation, input perturbation stress tests, SHAP attribution, and MCDA-based selection into a single pipeline, this work offers exactly such a recipe — one that other forecasting teams can adapt to their own rivers. The message for the emerging generation of AI-powered flood warning systems is clear: the models that save lives are not the ones that predict rivers best on average, but the ones that refuse to be caught off guard by the peak.

Subject of Research: Peak-oriented one-day-ahead streamflow forecasting with hybrid deep learning models

Article Title: Peak-oriented one-day-ahead streamflow forecasting using hybrid deep learning: uncertainty quantification, SHAP-based interpretability, and MCDA-based model selection

Article References: Sedaghat Rad, M. R., Feizi, A., & Alizadeh, H. (2026). Peak-oriented one-day-ahead streamflow forecasting using hybrid deep learning: uncertainty quantification, SHAP-based interpretability, and MCDA-based model selection. Natural Hazards, 122(19), Article 631. https://doi.org/10.1007/s11069-026-08404-2

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08404-2

Keywords: streamflow forecasting, flood prediction, deep learning, CNN-RNN hybrid, uncertainty quantification, Monte Carlo Dropout, SHAP interpretability, multi-criteria decision analysis, peak flow, early-warning systems, hydrology, Iran

The problem the researchers set out to solve is one of the most stubborn in modern exploration geophysics. Geophysical signals from deep ore bodies arrive at the surface attenuated and weak, their low amplitudes easily drowned out by near-surface geological background. Conventional inversion techniques that translate raw electromagnetic, magnetic, and gravity measurements into three-dimensional models of the subsurface suffer from non-uniqueness: many different underground configurations can produce nearly identical surface readings. When the target is a subtle mineralized body rather than a large, conductive massive sulfide, the ambiguity becomes severe. Human interpreters and standard machine learning tools alike struggle to decide which faint blips in an inversion volume are genuine exploration targets and which are artifacts.

What makes the challenge doubly hard is the label problem. Machine learning classifiers typically need thousands of labeled examples to learn reliably, but in mineral exploration the only trustworthy labels come from expensive deep drilling. At Kalatongke, the research team had just 906 labeled samples drawn from boreholes to constrain a volume of 2,023,623 three-dimensional inversion voxels. That means less than half of one tenth of one percent of the data carried any ground truth. Worse, mineralized intervals are dramatically outnumbered by barren rock, producing a severe class imbalance that causes ordinary classifiers to simply predict the majority class everywhere and miss the ore entirely.

LSR-GAD, short for a label-scarce reliability-aware graph anomaly detection framework, attacks both problems at once. The core idea is to represent the three-dimensional inversion volume as a graph rather than a stack of independent pixels. Each inversion voxel becomes a node in the graph, carrying its electrical, magnetic, and gravity inversion properties as features. A graph sample and aggregate network, known as GraphSAGE, then encodes local spatial relationships by aggregating information from neighboring nodes. This neighborhood aggregation matters because ore bodies are spatially continuous objects, not random scatterings of anomalous voxels; a voxel surrounded by other weakly anomalous voxels is far more likely to be part of a real mineralized zone than an isolated outlier, and the graph structure lets the model learn that context directly.

To cope with the scarcity of labels, the framework combines the small number of labeled borehole samples with unlabeled anchor nodes, allowing the network to extract useful structure from the vast unlabeled portion of the data while still being anchored by what little ground truth exists. The class imbalance is handled through a weighted binary cross-entropy loss, which penalizes errors on the rare mineralized class more heavily than errors on the abundant barren class, effectively telling the model that missing an ore body is a far costlier mistake than a false alarm in barren rock. Synthetic experiments and ablation tests confirmed that each ingredient contributes: removing the graph structure, the sample weighting, or the regularization each degraded the continuity of predicted targets and increased missed detections under sparse-label conditions.

Perhaps the most consequential design choice is the explicit treatment of uncertainty. Rather than producing a single brittle prediction, the framework incorporates Monte Carlo dropout, a technique in which the network’s dropout layers remain active at inference time so that many slightly different versions of the model each contribute a prediction. From this ensemble of predictions, the method estimates predictive probability, standard deviation, and entropy for every voxel. High probability with low variance marks confident exploration targets; high probability with high variance flags regions where the model is guessing. This turns the output from a binary map into a risk-aware decision tool, letting exploration managers weigh where to spend scarce drilling budgets against quantified confidence rather than blind faith in a black box.

Scale presented one final engineering hurdle. A million-scale voxel volume cannot be pushed through a graph network in one piece, so the team developed an anchor-graph-based block-wise inference strategy that partitions the volume into manageable blocks for prediction while preserving the benefits of the graph representation. This makes the approach practical for real exploration datasets, which routinely span kilometers of depth and lateral extent at fine spatial resolution. The combination of graph-based spatial reasoning, semi-supervised learning, class-rebalancing, uncertainty estimation, and scalable inference is what allows LSR-GAD to function reliably in exactly the regime where conventional supervised methods collapse: deep, subtle targets with almost no labels.

The Kalatongke case study demonstrates the framework on a genuinely difficult target. Kalatongke is a magmatic copper-nickel sulfide district hosted in the Central Asian Orogenic Belt, a vast accretionary collage where Permian mafic intrusions carried sulfide melts rich in copper and nickel. Known ore bodies have been extensively drilled, but the next generation of discoveries lies deeper and subtler, in concealed intrusions whose geophysical signatures are faint and entangled with background geology. The team integrated electrical, magnetic, and gravity inversion models, letting the classifier exploit the fact that a genuine mineralized body should leave coherent, mutually consistent traces across multiple physical properties rather than an artifact confined to a single dataset.

Trained on the 906 labeled samples, the model generated a full probability and uncertainty map of the survey volume. The stand-out result is the high-probability anomaly near line 805 at approximately 1.0 kilometer depth, which corresponds spatially to the gabbroic interval encountered in borehole ZK2017-6. Because that agreement was achieved with a tiny training set, it suggests the framework can generalize from limited ground truth in ways that matter for greenfield exploration, where deep drilling data simply does not yet exist. The authors position the output explicitly as a risk-management tool: probability and uncertainty layers guide drill targeting, and each new borehole feeds back into the training set, progressively sharpening the model as a district matures.

Beyond one deposit, the study points toward a broader shift in how geoscience handles sparse, expensive labels. Deep learning is spreading rapidly through geophysics, but most successful applications sit in data-rich domains; mineral exploration, with its handful of boreholes against millions of unlabeled voxels, has remained a hostile environment for standard supervised learning. By fusing multi-source geophysical data on a graph, rebalancing the loss, and quantifying its own uncertainty, LSR-GAD offers a reusable template for intelligent exploration of complex concealed ore bodies, whether the target is nickel-copper sulfide, rare earths, or energy-critical metals. As demand for critical minerals accelerates worldwide, tools that can squeeze reliable guidance from minimal ground truth may become as essential to explorers as the magnetometer itself.

Subject of Research: A label-scarce reliability-aware graph anomaly detection framework for identifying deep subtle mineralization anomalies from multi-source geophysical inversion data, applied to the Kalatongke Cu-Ni deposit in the Central Asian Orogenic Belt.

Article Title: LSR-GAD for Reliable Detection of Deep Subtle Mineralization Anomalies under Label Scarcity: A Case Study of the Kalatongke Deposit in the Central Asian Orogenic Belt

Article References: Lv, P., Zhou, N., Chen, W., & Han, S. (2026). LSR-GAD for Reliable Detection of Deep Subtle Mineralization Anomalies under Label Scarcity: A Case Study of the Kalatongke Deposit in the Central Asian Orogenic Belt. Natural Resources Research. https://doi.org/10.1007/s11053-026-10765-1

Image Credits: AI Generated

DOI: 10.1007/s11053-026-10765-1

Keywords: graph neural network, mineral exploration, label scarcity, geophysical inversion, Kalatongke deposit, uncertainty quantification, Monte Carlo dropout, GraphSAGE, multi-source data fusion, copper-nickel sulfide, Central Asian Orogenic Belt, machine learning